08.12.01 · stat-mech / linear-response

Fluctuation-dissipation theorem (Landau-Callen-Welton)

shipped3 tiersLean: none

Anchor (Master): Landau-Lifshitz, *Statistical Physics, Part 1*, 3rd ed. (Pergamon, 1980), Vol. 5, §§124-126 (canonical FDT derivation); Callen-Welton, *Phys. Rev.* 83, 34 (1951); Kubo, *J. Phys. Soc. Jpn.* 12, 570 (1957); Kubo-Toda-Hashitsume, *Statistical Physics II: Nonequilibrium Statistical Mechanics* (Springer, 1985), Chs. 2-4; Zwanzig, *Nonequilibrium Statistical Mechanics* (Oxford, 2001), Chs. 1-2; Bratteli-Robinson, *Operator Algebras and Quantum Statistical Mechanics 2*, 2nd ed. (Springer, 1997), §5.3 (KMS condition)

Intuition Beginner

A resistor sitting on your bench at room temperature does nothing visible. Connect a sensitive voltmeter and the screen shows a noisy signal: the voltage across the resistor jitters around zero with no battery attached and no current flowing. The jitter is real, not measurement error, and it has a size set entirely by the temperature of the resistor and the value of the resistance. The hotter the resistor or the larger the resistance, the larger the jitter. This noise is a window into the random thermal motion of the electrons inside.

The same resistor also dissipates energy when you do drive it. Push a current through with a battery and the resistor heats up. The rate of heating is set by the resistance: a larger resistance dissipates more power for the same current. The deep fact is that the spontaneous voltage jitter at zero current and the dissipated heat under driven current are two sides of one coin. Knowing how much heat the resistor dissipates tells you exactly how much noise it makes when left alone, and vice versa.

This is the fluctuation-dissipation theorem. For any system in thermal equilibrium, the spontaneous fluctuations of any observable and the energy the system dissipates when that observable is driven are the same quantity in two different costumes. The theorem is what lets you predict noise spectra from response measurements, predict response from noise, and connect every Brownian-motion problem to every linear-response problem. The wonderfully general phrase you sometimes hear is that nature plays the same game whether you are watching it relax or shoving it around.

Visual Beginner

Picture a resistor on a benchtop. Above it, two graphs side by side. On the left, a noisy voltage trace wandering around zero in real time, with the height of the wandering set by the temperature of the resistor. On the right, the same resistor driven by an external voltage at some frequency: the input is a smooth sine wave, the output is a smoothed sine wave at the same frequency but smaller, with the size of the loss set by the resistance. An arrow connects the two pictures: the area under the noise spectrum on the left equals four times the temperature times the resistance, the same resistance that controls the dissipation on the right.

The picture captures the structural claim: spontaneous noise on the left and driven response on the right are not separate phenomena. They share a single underlying object, the response function of the resistor, evaluated at thermal equilibrium. Replace the resistor with a Brownian particle in a fluid, a magnet in a small magnetic field, a polymer in a solvent: the same picture holds, with the appropriate observable and the appropriate susceptibility.

Worked example Beginner

Estimate the voltage noise of a $1 k Ω$ resistor at room temperature in a bandwidth of $1 Hz$ to $10 kHz$ . The Johnson-Nyquist formula says the mean-square voltage fluctuation across the resistor is $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ , where $k_{B}$ is the Boltzmann constant, $T$ is the temperature in kelvin, $R$ is the resistance in ohms, and $Δ ν$ is the bandwidth in hertz.

Step 1. Plug in the numbers. The Boltzmann constant is $k_{B} = 1.38 \times 1 0^{- 23} J/K$ . Room temperature is $T = 300 K$ . The resistance is $R = 1000 ohms$ . The bandwidth is $Δ ν = 10000 - 1 = 9999 Hz$ , which to the accuracy of this estimate is just $Δ ν = 1 0^{4} Hz$ .

Step 2. Multiply. The product $4 k_{B} T R$ equals $4 \times 1.38 \times 1 0^{- 23} \times 300 \times 1 0^{3} = 1.656 \times 1 0^{- 17} V^{2} / Hz$ . Then multiply by the bandwidth $1 0^{4} Hz$ to get the mean-square voltage: $⟨ V^{2} ⟩ = 1.656 \times 1 0^{- 13} V^{2}$ .

Step 3. Take the square root for the root-mean-square voltage: $V_{rms} = ⟨ V^{2} ⟩ \approx 4 \times 1 0^{- 7} V = 0.4 microvolts$ . So a one-kilohm resistor at room temperature, measured in a 10 kHz bandwidth, has a noise floor of about half a microvolt.

Step 4. Check the dependence. Double the temperature to $600 K$ and the mean-square noise doubles, the rms noise grows by a factor of $2$ . Increase the resistance to a megohm ( $1 0^{6}$ ohms) and the mean-square noise grows by a factor of $1000$ , the rms noise by $1000 \approx 32$ . The noise scales as the square root of the temperature times the resistance times the bandwidth.

Step 5. Where does the noise come from? The electrons inside the resistor are in constant thermal motion. Each scattering event off a lattice ion deflects an electron by a tiny random amount. The accumulated random walk of all the electrons produces a fluctuating voltage at the terminals. The size of the fluctuation is set by how strongly the electrons couple to the lattice, which is the same coupling that controls the resistance. Larger resistance equals larger coupling equals larger noise.

What this tells us: the fluctuation-dissipation theorem is not a metaphor. It is a quantitative formula that predicts the noise of a resistor from its resistance and temperature, with no adjustable parameters. The same formula applies to any equilibrium system: read the response function from a driven measurement, plug it in, and the noise spectrum comes out. The reverse direction works too: measure the noise spectrum and read off the response function.

Check your understanding Beginner

Exercise (easy, multiple-choice).

The fluctuation-dissipation theorem relates which two quantities?

A. The position and momentum of a particle
B. The temperature and entropy of a gas
C. The spontaneous fluctuations of an observable and the dissipative response to a driving field
D. The kinetic and potential energy of a harmonic oscillator

Hint

The theorem identifies the size of equilibrium noise with the dissipative part of the susceptibility that controls how the system absorbs energy from an external drive.

Answer

C. The spontaneous fluctuations of an observable and the dissipative response to a driving field. The theorem says that the equilibrium noise spectrum of an observable and the dissipative (imaginary) part of its response function are the same quantity, related by a temperature-dependent factor.

Formal definition Intermediate+

Let $H$ be a separable Hilbert space and $H$ a self-adjoint Hamiltonian on $H$ . The canonical equilibrium state at inverse temperature $β = 1/ (k_{B} T)$ is the density operator $ρ_{e q} = e^{- β H} / Z$ with $Z = Tr (e^{- β H})$ , defined when the trace converges. The equilibrium expectation of an observable $A$ is $⟨ A ⟩ = Tr (ρ_{e q} A)$ . The Heisenberg evolution of $A$ is $A (t) = e^{i H t /ℏ} A e^{- i H t /ℏ}$ .

Let $A$ and $B$ be observables and let the system be coupled to a time-dependent external field $f (t)$ that couples to $B$ , modifying the Hamiltonian to $H (t) = H - f (t) B$ . The generalised susceptibility is the linear-response coefficient relating the deviation of $⟨ A ⟩$ from equilibrium to the applied field. To first order in $f$ , $$ \delta \langle A(t) \rangle = \int_{-\infty}^{t} \chi_{AB}(t - t') f(t') , dt', $$ where the retarded response function (Kubo formula) is $$ \chi_{AB}(t) = -\frac{i}{\hbar} \theta(t) , \langle [A(t), B(0)] \rangle_{eq}, $$ with $θ (t)$ the Heaviside step function and $[A, B] = A B - B A$ the commutator. The Fourier transform $χ_{A B} (ω) = \int e^{iω t} χ_{A B} (t) d t$ is the frequency-domain susceptibility, an analytic function in the upper half plane by the retarded support of $χ_{A B} (t)$ .

The equilibrium correlation function of $A$ and $B$ is $$ C_{AB}(t) := \langle A(t) B(0) \rangle_{eq} = \mathrm{Tr}\bigl(\rho_{eq} A(t) B\bigr). $$ The spectral function is the Fourier transform $$ S_{AB}(\omega) := \int_{-\infty}^{\infty} e^{i\omega t} , C_{AB}(t), dt. $$

The defining relation that ties $χ_{A B}$ and $S_{A B}$ together is the Kubo-Martin-Schwinger (KMS) condition, which holds on any thermal equilibrium state: $$ \langle A(t) B(0) \rangle_{eq} = \langle B(0) A(t + i\beta\hbar) \rangle_{eq}. $$ The KMS condition follows from the cyclicity of the trace and the analyticity of $e^{- β H} A (z) e^{β H}$ in the strip $0 \leq Im z \leq β ℏ$ for $A$ in a dense $*$ -subalgebra of bounded operators. It is the abstract characterisation of a thermal equilibrium state on a general $C^{*}$ -dynamical system ^{[Bratteli-Robinson §5.3]}.

The symmetrised and antisymmetrised correlations are $$ S^+{AB}(\omega) := \frac{1}{2}\bigl(S{AB}(\omega) + S_{BA}(-\omega)\bigr), \qquad \chi''{AB}(\omega) := \mathrm{Im}, \chi{AB}(\omega). $$ The fluctuation-dissipation theorem is the identity $$ \chi''{AB}(\omega) = \frac{1}{2\hbar}\bigl(1 - e^{-\beta\hbar\omega}\bigr), S{AB}(\omega), $$ equivalently $$ S_{AB}(\omega) = \frac{2\hbar}{1 - e^{-\beta\hbar\omega}}, \chi''{AB}(\omega). $$ In the classical limit $ℏ ω ≪ k_{B} T$ , expand $1 - e^{- β ℏ ω} \approx β ℏ ω$ and the identity reduces to the classical FDT $$ S{AB}(\omega) = \frac{2 k_B T}{\omega}, \chi''_{AB}(\omega). $$

Counterexamples to common slips

The factor $1 - e^{- β ℏ ω}$ is asymmetric in $ω$ , reflecting the asymmetry between absorption and emission in a thermal system. Replacing it with $coth (β ℏ ω /2)$ symmetrises the formula and yields the symmetrised FDT $S_{A B}^{+} (ω) = ℏ coth (β ℏ ω /2) χ_{A B}^{''} (ω)$ . The two forms are equivalent and a switch between them must keep track of the symmetrisation convention.
The retarded response function $χ_{A B} (t)$ vanishes for $t < 0$ by causality. The advanced response $χ_{A B}^{a d v} (t) = (i /ℏ) θ (- t) ⟨[A (t), B (0)]⟩$ is the time-reversed object and has support $t \leq 0$ . The FDT relates the equilibrium correlator to the imaginary part of the retarded susceptibility specifically; the advanced version differs by a sign.
The classical limit $ℏ \to 0$ is taken with $β$ held fixed and the argument $β ℏ ω$ sent to zero. Holding $β ℏ ω$ fixed while sending $ℏ \to 0$ samples high-frequency quantum effects (zero-point fluctuations) and does not give the classical answer. The dimensionless small parameter is $β ℏ ω = ℏ ω / (k_{B} T)$ .
The susceptibility $χ_{A B} (ω)$ is not in general symmetric in $(A, B)$ . Onsager reciprocity gives a refined symmetry: $χ_{A B} (ω) = ϵ_{A} ϵ_{B} χ_{B A} (ω)$ , where $ϵ_{A}, ϵ_{B} = \pm 1$ are the parities of $A$ and $B$ under time reversal. This is the FDT-compatible refinement of the naive symmetry expectation.

Key derivation Intermediate+

Theorem (Fluctuation-dissipation theorem; Callen-Welton 1951 ^{[Callen-Welton 1951]}, Kubo 1957 ^{[Kubo 1957]}). Let $A$ and $B$ be observables on a quantum system in thermal equilibrium at inverse temperature $β$ , and let $χ_{A B} (ω)$ and $S_{A B} (ω)$ be the retarded susceptibility and the spectral function defined above. Then $$ \mathrm{Im}, \chi_{AB}(\omega) = \frac{1}{2\hbar}\bigl(1 - e^{-\beta\hbar\omega}\bigr), S_{AB}(\omega). $$ Equivalently, $S_{A B} (ω) = (2ℏ/ (1 - e^{- β ℏ ω})) Im χ_{A B} (ω)$ . In the classical limit $ℏ ω ≪ k_{B} T$ , $S_{A B} (ω) = 2 k_{B} T \cdot Im χ_{A B} (ω) / ω$ .

Proof. Begin with the spectral representation. Write the canonical state as $ρ_{e q} = Z^{- 1} \sum_{n} e^{- β E_{n}} ∣ n ⟩ ⟨ n ∣$ , where $H ∣ n ⟩ = E_{n} ∣ n ⟩$ is the spectral decomposition. Then $$ C_{AB}(t) = \langle A(t) B(0)\rangle = \frac{1}{Z} \sum_{n, m} e^{-\beta E_n} e^{i (E_n - E_m) t / \hbar} A_{nm} B_{mn}, $$ where $A_{nm} = ⟨ n ∣ A ∣ m ⟩$ and $B_{mn} = ⟨ m ∣ B ∣ n ⟩$ . Fourier transform in $t$ : $$ S_{AB}(\omega) = \frac{2\pi}{Z} \sum_{n,m} e^{-\beta E_n} A_{nm} B_{mn}, \delta\bigl(\omega - (E_m - E_n)/\hbar\bigr). $$ Similarly, $$ C_{BA}(-t) = \langle B(-t) A(0)\rangle = \frac{1}{Z} \sum_{n,m} e^{-\beta E_n} e^{-i (E_n - E_m) t / \hbar} B_{nm} A_{mn}, $$ so $$ S_{BA}(-\omega) = \int e^{-i\omega t} C_{BA}(-t),dt = \frac{2\pi}{Z} \sum_{n,m} e^{-\beta E_n} B_{nm} A_{mn}, \delta\bigl(-\omega + (E_n - E_m)/\hbar\bigr). $$ Relabel $n \leftrightarrow m$ in the second expression: $$ S_{BA}(-\omega) = \frac{2\pi}{Z} \sum_{n,m} e^{-\beta E_m} B_{mn} A_{nm}, \delta\bigl((E_m - E_n)/\hbar - \omega\bigr) = e^{-\beta\hbar\omega} S_{AB}(\omega), $$ where the last equality uses $E_{m} = E_{n} + ℏ ω$ on the support of the delta function to pull out the factor $e^{- β (E_{m} - E_{n})} = e^{- β ℏ ω}$ . This is the detailed-balance relation for spectral functions: $$ S_{BA}(-\omega) = e^{-\beta\hbar\omega} S_{AB}(\omega). \tag{ $*$ } $$ It is equivalent to the KMS condition in Fourier space.

Now compute the response function. From the Kubo formula $$ \chi_{AB}(t) = -\frac{i}{\hbar} \theta(t) \bigl(C_{AB}(t) - C_{BA}(-t)\bigr). $$ Fourier transform: $$ \chi_{AB}(\omega) = -\frac{i}{\hbar} \int_0^\infty e^{i\omega t}\bigl(C_{AB}(t) - C_{BA}(-t)\bigr),dt. $$ Take the imaginary part. Using $Im χ_{A B} (ω) = - (1/2 i) (χ_{A B} (ω) - χ_{A B} (ω)^{*})$ and the reality of $C_{A B} (t)^{*} = ⟨ B (0) A (t)⟩ = C_{B A} (- t)$ (which follows from $A^{†} = A$ , $B^{†} = B$ for hermitian observables and the cyclicity of the trace), a direct computation gives $$ \mathrm{Im}, \chi_{AB}(\omega) = \frac{1}{2\hbar} \bigl(S_{AB}(\omega) - S_{BA}(-\omega)\bigr). $$ Substituting the detailed-balance identity ( $*$ ), $$ \mathrm{Im}, \chi_{AB}(\omega) = \frac{1}{2\hbar}\bigl(S_{AB}(\omega) - e^{-\beta\hbar\omega} S_{AB}(\omega)\bigr) = \frac{1}{2\hbar}\bigl(1 - e^{-\beta\hbar\omega}\bigr) S_{AB}(\omega). $$ This is the quantum FDT. Solving for $S_{A B} (ω)$ : $$ S_{AB}(\omega) = \frac{2\hbar}{1 - e^{-\beta\hbar\omega}}, \mathrm{Im}, \chi_{AB}(\omega). $$ In the classical limit $β ℏ ω \to 0$ , expand $1 - e^{- β ℏ ω} = β ℏ ω + O ((β ℏ ω)^{2})$ , hence $$ S_{AB}(\omega) \approx \frac{2\hbar}{\beta\hbar\omega} \mathrm{Im}, \chi_{AB}(\omega) = \frac{2 k_B T}{\omega}, \mathrm{Im}, \chi_{AB}(\omega). $$ $□$

Bridge. The fluctuation-dissipation theorem builds toward every linear-response computation in equilibrium statistical mechanics, and the same detailed-balance principle appears again in 08.10.02 as the reversibility condition that selects the Gibbs distribution as the equilibrium of the Fokker-Planck flow. The foundational reason the identity holds is exactly the KMS condition on the equilibrium state: cyclicity of the canonical trace plus the analyticity of imaginary-time evolution in the strip $0 \leq Im t \leq β ℏ$ gives the asymmetry $S_{B A} (- ω) = e^{- β ℏ ω} S_{A B} (ω)$ , and combining this with the commutator structure of the Kubo response function produces the FDT. This is exactly the structural identification that the Bratteli-Robinson C*-algebraic framework upgrades to a definition: a state on a $C^{*}$ -dynamical system is thermal at inverse temperature $β$ precisely when it satisfies the KMS condition. The central insight is that the FDT identifies the imaginary part of the susceptibility with the equilibrium fluctuation spectrum modulated by the Bose-Einstein factor $(1 - e^{- β ℏ ω})^{- 1}$ . The bridge is that this identification generalises beyond the canonical ensemble. Putting these together, one FDT framework produces every linear-response coefficient (conductivities, susceptibilities, viscosities) from equilibrium correlation functions, and identifies them with the corresponding fluctuation spectra.

Two refinements complete the picture. Replacing the canonical ensemble with a grand canonical ensemble gives a multi-observable FDT for currents driven by chemical-potential differences. Replacing the linear coupling $- f (t) B$ with a non-equilibrium initial condition gives the Onsager regression hypothesis, that equilibrium fluctuations relax on the same timescale as non-equilibrium responses. The classical limit reduces to the Einstein relation $γ D = k_{B} T$ on a Brownian particle and to the Johnson-Nyquist formula $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ on a resistor, the two historical originators of the modern theorem.

Exercises Intermediate+

Exercise 4 (medium, symbolic).

Derive the Johnson-Nyquist formula from the classical FDT. Model the resistor as a black box with susceptibility $χ_{I I} (ω) = 1/ (R - iω L)$ relating current to applied voltage (an admittance with possible inductance), and compute $⟨ V^{2} ⟩$ in the limit of vanishing inductance $L \to 0$ and a flat bandwidth $Δ ν$ .

Hint

Use the classical FDT $S_{V V} (ω) = (2 k_{B} T / ω) Im χ_{V V} (ω)$ where $χ_{V V} (ω) = 1/ Y (ω)^{*} = R - iω L$ for an ohmic resistor, then integrate $S_{V V} (ω) / (2 π)$ over the bandwidth $Δ ν$ .

Answer

The impedance of an ohmic resistor in the limit $L \to 0$ is $Z (ω) = R$ , real and frequency-independent. The voltage susceptibility (response of voltage to applied current source) is $χ_{V V} (ω) = R$ , but the relevant FDT involves the response of current to applied voltage, whose imaginary part vanishes for a pure resistor in the inductance-free limit. The proper formulation uses the voltage-noise spectral density at fixed temperature: by the classical FDT applied to the current-current correlation, the equilibrium voltage spectrum is $S_{V V} (ω) = 2 k_{B} T R$ (frequency-independent for an ideal resistor). Integrating over the bandwidth $Δ ν$ in single-sided convention (positive frequencies, factor of $2$ from folding the two-sided spectrum), the total mean-square voltage is $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ . This is exactly the Nyquist 1928 formula derived from a first-principles FDT computation. Rubric: full credit for $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ with the FDT-derived spectrum.

Exercise 5 (medium, short-answer).

State the KMS condition and explain why it captures thermal equilibrium. Show how the KMS condition reduces to the classical Boltzmann distribution in the limit $ℏ \to 0$ .

Hint

The KMS condition is $⟨ A (t) B (0)⟩ = ⟨ B (0) A (t + i β ℏ)⟩$ . In the classical limit, the analytic continuation $t \to t + i β ℏ$ acts as the identity on commuting observables, and the condition becomes a statement about Liouville-time-translated correlations.

Answer

The KMS condition states that for any pair of bounded observables $A, B$ and any real time $t$ , the equilibrium expectation satisfies $⟨ A (t) B (0) ⟩_{e q} = ⟨ B (0) A (t + i β ℏ) ⟩_{e q}$ , where the right-hand side uses the analytic continuation of $A (t)$ to imaginary times in the strip $Im t \in [0, β ℏ]$ . The condition is equivalent to the canonical density operator $ρ_{e q} = Z^{- 1} e^{- β H}$ when one starts from the trace expression and uses cyclicity. The Bratteli-Robinson program promotes KMS to a definition: a state on a $C^{*}$ -dynamical system $(A, σ_{t})$ is a thermal equilibrium state at inverse temperature $β$ if and only if it satisfies the KMS condition with the dynamics $σ_{t}$ as time translation. In the classical limit $ℏ \to 0$ , observables commute and the analytic continuation $t \to t + i β ℏ$ collapses (the imaginary shift acts as the identity on commuting variables). The KMS condition reduces to $⟨ A B ⟩ = ⟨ B A ⟩ = \int A (x) B (x) e^{- β H (x)} / Z d x$ , the classical Boltzmann distribution as the equilibrium measure on phase space. Rubric: full credit for the KMS statement plus the classical reduction.

Exercise 6 (medium, symbolic).

For a harmonic oscillator of mass $m$ and frequency $ω_{0}$ , the position susceptibility is $χ_{xx} (ω) = 1/ (m (ω_{0}^{2} - ω^{2} - i η ω))$ with $η > 0$ a small damping (added to make the response retarded). Compute the equilibrium position fluctuation $⟨ x^{2} ⟩$ at temperature $T$ using the classical FDT and Parseval's theorem.

Hint

Use $⟨ x^{2} ⟩ = \int_{- \infty}^{\infty} S_{xx} (ω) d ω / (2 π)$ and the classical FDT $S_{xx} (ω) = (2 k_{B} T / ω) Im χ_{xx} (ω)$ . The integral can be evaluated by residues or by recognising the spectral density as a sharply peaked Lorentzian at $ω = \pm ω_{0}$ .

Answer

Compute $Im χ_{xx} (ω) = η ω / (m [(ω_{0}^{2} - ω^{2})^{2} + η^{2} ω^{2}])$ . The classical FDT then gives $S_{xx} (ω) = (2 k_{B} T / ω) \cdot η ω / (m [(ω_{0}^{2} - ω^{2})^{2} + η^{2} ω^{2}]) = 2 k_{B} T η / (m [(ω_{0}^{2} - ω^{2})^{2} + η^{2} ω^{2}])$ . The mean-square position is $$ \langle x^2\rangle = \int_{-\infty}^\infty \frac{d\omega}{2\pi} \frac{2 k_B T \eta}{m[(\omega_0^2 - \omega^2)^2 + \eta^2 \omega^2]}. $$ Evaluating by contour integration (close in the upper half plane and collect residues at the poles $ω = \pm ω_{0} + i η /2$ to leading order in small $η$ ), the integral gives $⟨ x^{2} ⟩ = k_{B} T / (m ω_{0}^{2})$ , independent of the damping $η$ . This is the equipartition result for a quadratic mode in a harmonic potential at temperature $T$ . Rubric: full credit for $⟨ x^{2} ⟩ = k_{B} T / (m ω_{0}^{2})$ with the FDT-based derivation.

Exercise 7 (hard, short-answer).

State and prove Onsager's reciprocity theorem in the form $χ_{A B} (ω) = ϵ_{A} ϵ_{B} χ_{B A} (ω)$ , where $ϵ_{A}, ϵ_{B} = \pm 1$ are the parities of $A$ and $B$ under time reversal. Explain why this is the FDT-compatible refinement of the naive symmetry expectation $χ_{A B} = χ_{B A}$ .

Hint

Use time-reversal symmetry of the equilibrium state ( $Θ ρ_{e q} Θ^{- 1} = ρ_{e q}$ with $Θ$ the antiunitary time-reversal operator) on the spectral representation of $χ_{A B} (ω)$ , and track the parity factors $ϵ_{A}, ϵ_{B}$ defined by $Θ A Θ^{- 1} = ϵ_{A} A$ .

Answer

The setup. Let $Θ$ be the antiunitary time-reversal operator, satisfying $Θ i Θ^{- 1} = - i$ and $Θ H Θ^{- 1} = H$ (the Hamiltonian is time-reversal-symmetric). The equilibrium density operator is invariant: $Θ ρ_{e q} Θ^{- 1} = ρ_{e q}$ . Observables transform with definite parity: $Θ A Θ^{- 1} = ϵ_{A} A$ , $Θ B Θ^{- 1} = ϵ_{B} B$ , with $ϵ_{A}, ϵ_{B} = \pm 1$ (positions are even, momenta and magnetic moments are odd). Then $Θ A (t) Θ^{- 1} = ϵ_{A} A (- t)$ because $Θ e^{i H t /ℏ} Θ^{- 1} = e^{- i H t /ℏ}$ .

The proof. The retarded susceptibility is $χ_{A B} (t) = - (i /ℏ) θ (t) ⟨[A (t), B (0)]⟩$ . Using cyclicity and the antiunitarity of $Θ$ , $⟨[A (t), B (0)]⟩ = Tr (ρ_{e q} [A (t), B]) = Tr (Θ ρ_{e q} [A (t), B] Θ^{- 1})^{*} = Tr (ρ_{e q} [Θ A (t) Θ^{- 1}, Θ B Θ^{- 1}])^{*} = ϵ_{A} ϵ_{B} ⟨[A (- t), B] ⟩^{*}$ . The complex conjugate of an expectation of a commutator is minus the expectation of the swapped commutator (the commutator is anti-hermitian for hermitian observables), so $⟨[A (- t), B] ⟩^{*} = - ⟨[B, A (- t)]⟩ = ⟨[A (- t), B]⟩$ flipping signs once more after using hermiticity. Tracking the signs carefully: $χ_{A B} (t) = ϵ_{A} ϵ_{B} χ_{B A} (t)$ in the time domain (the swap $A \leftrightarrow B$ and $t \to - t$ undo each other under time-reversal symmetry), and Fourier transforming gives the frequency-domain reciprocity $χ_{A B} (ω) = ϵ_{A} ϵ_{B} χ_{B A} (ω)$ .

Why it is the FDT-compatible refinement. The naive expectation $χ_{A B} = χ_{B A}$ would hold for commuting observables in a classical system, but quantum observables generically do not commute, and the FDT involves a commutator structure. The parity factors $ϵ_{A} ϵ_{B}$ correct for the antisymmetry of the time-reversal operation on odd observables, ensuring compatibility with the KMS condition. In matrix language, the susceptibility tensor $χ_{ij} (ω)$ in linear irreversible thermodynamics decomposes as $χ_{ij} = χ_{ij}^{s} + χ_{ij}^{a}$ with $χ_{ij}^{s} = χ_{j i}^{s}$ symmetric and $χ_{ij}^{a} = - χ_{j i}^{a}$ antisymmetric (the latter present only when at least one of $A_{i}, A_{j}$ is time-reversal-odd, like a magnetic moment). The Onsager 1931 Phys. Rev. result $L_{ij} = L_{j i}$ for kinetic coefficients is the static $ω \to 0$ limit of this reciprocity. Rubric: full credit for the statement, the time-reversal proof, and the FDT compatibility argument.

Exercise 8 (hard, short-answer).

State Kubo's second fluctuation-dissipation theorem for a generalised Langevin equation, and show that it follows from the FDT applied to the noise force in a projection-operator (Mori-Zwanziger) derivation.

Hint

The generalised Langevin equation has the form $m \overset{v}{˙} (t) = - \int_{0}^{t} γ (t - s) v (s) d s + F (t)$ with $γ$ a memory friction kernel and $F$ a random force. Kubo's second FDT relates $⟨ F (t) F (0)⟩$ to $γ (t)$ via the temperature.

Answer

The statement. For a generalised Langevin equation $$ m \dot v(t) = -\int_0^t \gamma(t-s), v(s), ds + F(t), $$ where $v (t)$ is the velocity of a particle in contact with a thermal bath, $γ (t)$ is a memory friction kernel, and $F (t)$ is a stationary Gaussian random force with $⟨ F (t)⟩ = 0$ , the second FDT states $$ \langle F(t) F(0)\rangle = k_B T \cdot \gamma(t). $$ The noise force autocorrelation is the temperature times the friction memory kernel.

The derivation. Apply the Mori-Zwanziger projection-operator formalism to a Hamiltonian system with the particle weakly coupled to a bath. Project the full equation of motion for the velocity onto the orthogonal complement of the velocity in the equilibrium $L^{2}$ inner product. The resulting equation for $v (t)$ has the generalised Langevin form, with $γ (t) = (β / m) ⟨ F (t) F (0)⟩$ identified as the autocorrelation of the projected force $F (t) = e^{i tQ L Q} (Q Lv) ∣_{0}$ , where $Q$ projects out the velocity subspace and $L$ is the Liouville operator. Solving for $⟨ F (t) F (0)⟩$ gives the second FDT. The first FDT (Callen-Welton) is recovered as the $ω = 0$ limit when the system is in linear response to an external force coupled to the position. Rubric: full credit for the statement of the second FDT plus the projection-operator derivation.

Advanced results Master

Theorem (Onsager reciprocity; Onsager 1931 ^{[Onsager 1931]}). Let $A_{1}, \dots, A_{n}$ be observables on a quantum system in thermal equilibrium, each with definite time-reversal parity $ϵ_{i} = \pm 1$ . The susceptibility tensor $χ_{ij} (ω)$ satisfies $$ \chi_{ij}(\omega) = \epsilon_i \epsilon_j, \chi_{ji}(\omega). $$ In the static limit $ω \to 0$ , the kinetic coefficients $L_{ij} = lim_{ω \to 0} χ_{ij} (ω) / (iω)$ satisfy $L_{ij} = ϵ_{i} ϵ_{j} L_{j i}$ , recovering Onsager's classical reciprocity for time-reversal-even observables.

The reciprocity follows from time-reversal symmetry of the equilibrium state, as worked out in Exercise 7. The theorem decomposes the susceptibility tensor into a symmetric part (the dissipative response) and an antisymmetric part (present when at least one observable is time-reversal-odd, the canonical example being a Hall conductivity in a magnetic field). The FDT applies separately to each part: the symmetric part is what the dissipation-fluctuation identity controls; the antisymmetric part is the non-dissipative gyroscopic / reactive response.

Theorem (Kubo's second fluctuation-dissipation theorem; Kubo 1966 ^{[Kubo 1966]}). Let a particle of mass $m$ obey a generalised Langevin equation $$ m, \dot v(t) = -\int_0^t \gamma(t - s), v(s), ds + F(t), $$ with $γ$ a memory friction kernel and $F$ a stationary Gaussian random force with $⟨ F (t)⟩ = 0$ . Then in equilibrium at temperature $T$ , $$ \langle F(t) F(0)\rangle = k_B T, \gamma(t). $$ The noise autocorrelation is the temperature times the memory friction kernel.

This is the second FDT, distinguished from the first (Callen-Welton) FDT which relates the susceptibility to the equilibrium correlator. The second FDT relates the noise kernel of an effective stochastic dynamics to the friction kernel of the same dynamics. The two FDTs are equivalent statements when applied to a system in linear response to an external force: the first determines the response from the equilibrium correlator, the second determines the equilibrium correlator from the noise.

Theorem (Jarzynski equality; Jarzynski 1997 ^{[Jarzynski 1997]}). Let a system be driven by a time-dependent protocol $λ (t)$ between two equilibrium states with free energies $F_{0}$ and $F_{1}$ , and let $W$ be the work done on the system over a single realisation. Then $$ \langle e^{-\beta W}\rangle = e^{-\beta \Delta F}, $$ where $Δ F = F_{1} - F_{0}$ and the average is over the ensemble of initial conditions drawn from the equilibrium distribution of $λ (0)$ and the stochastic realisations of the dynamics. By Jensen's inequality $⟨ W ⟩ \geq Δ F$ , recovering the second law as the linear-response consequence of the non-equilibrium Jarzynski identity.

The Jarzynski equality is an exact non-equilibrium identity that contains the FDT as its linear-response limit. Expanding $e^{- β W} = 1 - β W + (β W)^{2} /2 - \dots$ and using $⟨ W ⟩ = Δ F + \frac{β}{2} (⟨ W^{2} ⟩ - ⟨ W ⟩^{2})$ to second order in the protocol speed, one recovers the FDT for the work fluctuations in the slow-driving limit. The full Jarzynski equality is the resummation of the perturbative FDT series.

Theorem (Crooks fluctuation theorem; Crooks 1999 ^{[Crooks 1999]}). Let $P_{F} (W)$ be the probability density of work $W$ under a forward protocol $λ (t)$ from $λ (0)$ to $λ (τ)$ , and $P_{R} (W)$ the probability density of work under the time-reversed protocol $λ (τ - t)$ . Then $$ \frac{P_F(W)}{P_R(-W)} = e^{\beta(W - \Delta F)}. $$ The Crooks fluctuation theorem refines the Jarzynski equality and reduces to it on integrating $P_{F} (W) e^{- β W} = P_{R} (- W) e^{- β Δ F}$ over $W$ .

The Crooks theorem is a detailed-balance identity at the trajectory level. It encodes more information than the Jarzynski equality (which is its integrated form) and contains the FDT as a Gaussian-work approximation in the slow-driving limit. Crooks-Jarzynski identities open the FDT to non-equilibrium stochastic thermodynamics, where they organise the modern theory of small-system thermodynamics and single-molecule measurements.

Theorem (quantum FDT and detailed balance in Lindblad master equations). Let $ρ (t)$ evolve under a Lindblad master equation $$ \dot\rho = -\frac{i}{\hbar}[H, \rho] + \sum_k \bigl(L_k \rho L_k^\dagger - \tfrac{1}{2}{L_k^\dagger L_k, \rho}\bigr), $$ with jump operators $L_{k}$ satisfying the quantum detailed balance condition $L_{k}^{†} e^{- β H} = e^{- β H} L_{k^{'}}$ for some pairing $k \leftrightarrow k^{'}$ . Then the canonical state $ρ_{e q} = e^{- β H} / Z$ is a stationary solution, and the FDT $χ^{''} (ω) = (1/2ℏ) (1 - e^{- β ℏ ω}) S (ω)$ holds for the linear response of $ρ_{e q}$ to small perturbations.

The quantum detailed-balance condition on Lindblad operators is the open-system analogue of the KMS condition on a closed system. It selects the Gibbs state as the unique fixed point of the dissipative dynamics and ensures the FDT survives the open-system embedding. Markovian quantum master equations not satisfying quantum detailed balance can have steady states far from equilibrium and violate the FDT.

Synthesis. The fluctuation-dissipation theorem is the foundational reason equilibrium statistical mechanics has a linear-response theory at all. The central insight is that the spectrum of spontaneous equilibrium fluctuations and the dissipative part of the response to an external drive are the same quantity in two presentations, related by the Bose-Einstein factor $(1 - e^{- β ℏ ω})^{- 1}$ in the quantum case and by the Einstein factor $2 k_{B} T / ω$ in the classical case. Putting these together, every linear-response coefficient (electrical conductivity, magnetic susceptibility, viscosity, diffusion coefficient) has a Green-Kubo formula expressing it as the equilibrium time-integral of the corresponding current-current correlation function, and the spectral content of those correlation functions is what the FDT pins down. The bridge between the abstract identification and concrete predictions is the spectral representation of the equilibrium state: writing $ρ_{e q} = Z^{- 1} \sum_{n} e^{- β E_{n}} ∣ n ⟩ ⟨ n ∣$ and Fourier-decomposing $⟨ A (t) B (0)⟩$ in energy differences $E_{m} - E_{n}$ , the detailed-balance relation $S_{B A} (- ω) = e^{- β ℏ ω} S_{A B} (ω)$ is just the statement that the energy of the system after absorbing or emitting a quantum $ℏ ω$ differs by exactly $ℏ ω$ , and the Boltzmann weights track that difference. This is exactly the principle that appears again in 08.10.02 as the detailed-balance current $J_{e q} = 0$ selecting the Gibbs distribution as the reversible stationary solution of the Fokker-Planck equation. The FDT generalises to spectra of stochastic dynamics (second FDT) and to non-equilibrium thermodynamics (Jarzynski-Crooks identities), where it organises the modern theory of small-system thermodynamics and provides the inequalities recovering the second law as a linear-response consequence.

The duality identifies several pairings that look distinct at first inspection. The Einstein relation $γ D = k_{B} T$ between drag and diffusion, the Nyquist formula $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ between voltage noise and resistance, the Kubo formula $σ (ω) = \int_{0}^{\infty} e^{iω t} ⟨ J (t) J (0)⟩ d t / (k_{B} T)$ for the electrical conductivity, and the Green-Kubo expressions for transport coefficients are all special cases of the same FDT applied to different observables. The bridge is that they all compute one number — the imaginary part of a susceptibility — from one input — the equilibrium correlation function — via one identity, the Callen-Welton 1951 formula.

The framework also identifies what the FDT does not capture. Spontaneous symmetry breaking is invisible to the linear-response framework because the symmetry-breaking expectation value lives in the static $ω \to 0$ susceptibility as a pole (a soft mode), not in the dissipative response. Phase transitions are visible as divergences of the static susceptibility, and the FDT translates these into divergences of the equilibrium fluctuation correlation length, identifying critical opalescence in fluids with the divergence of the static dielectric susceptibility at the critical point. Far-from-equilibrium driven systems require Crooks-Jarzynski refinements: the FDT-violation ratio $X_{A B} (ω) := χ_{A B}^{''} (ω) / (ω S_{A B} (ω) / (2 k_{B} T_{eff}))$ provides an operational definition of effective temperature in glasses and aging systems, where a single $T$ fails to capture the relation between fluctuation and dissipation across all timescales.

Full proof set Master

Theorem (Fluctuation-dissipation theorem), proof. Given in the Intermediate-tier derivation: the spectral representation of the canonical state gives $S_{B A} (- ω) = e^{- β ℏ ω} S_{A B} (ω)$ via relabeling and the delta-function support, and combining this detailed-balance identity with $Im χ_{A B} (ω) = (1/2ℏ) (S_{A B} (ω) - S_{B A} (- ω))$ from the Kubo formula yields the Callen-Welton identity. The classical limit follows by expansion of $1 - e^{- β ℏ ω}$ to first order. $□$

Theorem (Onsager reciprocity), proof. The retarded susceptibility is $χ_{A B} (t) = - (i /ℏ) θ (t) ⟨[A (t), B (0)]⟩$ . Time-reversal symmetry of the equilibrium state gives $Θ ρ_{e q} Θ^{- 1} = ρ_{e q}$ , and observables of definite parity transform as $Θ A (t) Θ^{- 1} = ϵ_{A} A (- t)$ . Applying $Θ$ to the trace and using antiunitarity (complex conjugation in the trace), one obtains $$ \langle[A(t), B(0)]\rangle = \epsilon_A \epsilon_B \langle[B(0), A(-t)]\rangle = -\epsilon_A \epsilon_B \langle[A(-t), B(0)]\rangle. $$ The retarded support $θ (t)$ combines with the time reversal $t \to - t$ to give $χ_{A B} (t) = ϵ_{A} ϵ_{B} χ_{B A} (t)$ at the level of the full Fourier integral (the support flip swaps the retarded and advanced response, and complex conjugation in the trace flips the sign of the commutator expectation, with the two minus signs cancelling). Fourier transform: $χ_{A B} (ω) = ϵ_{A} ϵ_{B} χ_{B A} (ω)$ . $□$

Theorem (Kubo's second FDT), proof sketch. Apply the Mori-Zwanziger projection-operator method to the Heisenberg equation $\overset{v}{˙} (t) = (i /ℏ) [H, v]$ in a Hamiltonian system. Project onto the subspace orthogonal to $v$ in the equilibrium inner product $⟨ X, Y ⟩ = Tr (ρ_{e q} X^{†} Y)$ . The projected dynamics gives an exact generalised Langevin equation $$ m\dot v(t) = -\int_0^t \gamma(t-s) v(s), ds + F(t), $$ with the random force $F (t) = e^{i tQ L Q} (Q Lv) ∣_{t = 0}$ and the friction kernel $γ (t) = (β / m) ⟨ F (t) F (0)⟩$ , where $Q = 1 - P$ projects out the velocity subspace and $L$ is the Liouville operator $L X = (i /ℏ) [H, X]$ . Solving for $⟨ F (t) F (0)⟩$ in terms of $γ$ gives $⟨ F (t) F (0)⟩ = (m / β) γ (t) = k_{B} T γ (t)$ , the second FDT. The full derivation is in Zwanzig 2001 §1.6 ^{[Zwanzig 2001]}. $□$

Proposition (Einstein relation $γ D = k_{B} T$ from FDT). For a Brownian particle of mass $m$ in a viscous fluid with friction coefficient $γ$ , the diffusion coefficient defined by the Green-Kubo formula $D = \int_{0}^{\infty} ⟨ v (t) v (0)⟩ d t$ satisfies $γ D = k_{B} T$ .

Proof. The Langevin equation $m \overset{v}{˙} = - γ v + F (t)$ with $⟨ F (t) F (0)⟩ = 2 γ k_{B} T δ (t)$ (Markovian limit of Kubo's second FDT) is solved by $v (t) = v_{0} e^{- γ t / m} + (1/ m) \int_{0}^{t} e^{- γ (t - s) / m} F (s) d s$ . In equilibrium $⟨ v (0)^{2} ⟩ = k_{B} T / m$ by equipartition. The time-displaced correlation, averaging over noise realisations and equilibrium initial conditions, is $$ \langle v(t) v(0)\rangle = \frac{k_B T}{m} e^{-\gamma t/m}. $$ Integrate over $t \in [0, \infty)$ : $$ D = \int_0^\infty \frac{k_B T}{m} e^{-\gamma t/m},dt = \frac{k_B T}{m} \cdot \frac{m}{\gamma} = \frac{k_B T}{\gamma}. $$ Hence $γ D = k_{B} T$ . This is the Einstein 1905 ^{[Einstein 1905]} relation, the historical precursor of the FDT. $□$

Proposition (Johnson-Nyquist from FDT). A resistor with resistance $R$ at temperature $T$ has voltage fluctuations $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ in any frequency bandwidth $Δ ν$ .

Proof. Model the resistor as an ohmic element with current-voltage response $V = R I$ at all frequencies in the relevant range (frequency-independent impedance). The voltage susceptibility, the linear response of voltage to an injected current source, is $χ_{V V} (ω) = R$ , real and frequency-independent. The relevant FDT involves the imaginary part of the impedance-related susceptibility; for an ohmic resistor with $Z (ω) = R$ , the dissipation per unit frequency is $Re Z (ω) = R$ , and the classical FDT in the form of the Nyquist representation gives $$ S_{VV}(\omega) = 2 k_B T \cdot R \quad \text{(two-sided spectral density)}, $$ or equivalently $S_{V V}^{(1)} (ω) = 4 k_{B} T R$ in the single-sided convention. Integrating over a bandwidth $Δ ν$ in the single-sided convention, $$ \langle V^2\rangle = \int_{\text{band}} S_{VV}^{(1)}(\nu),d\nu = 4 k_B T R \cdot \Delta\nu. $$ This is the Nyquist 1928 ^{[Nyquist 1928]} formula, derived as a consequence of the classical FDT applied to the equilibrium fluctuations of the voltage observable. $□$

Proposition (Jarzynski reduces to FDT in linear response). In the limit of slow driving with infinitesimal protocol variation $λ (t) = λ_{0} + ϵ \tilde{λ} (t)$ and work $W = ϵ W_{1} + ϵ^{2} W_{2} + O (ϵ^{3})$ , the Jarzynski equality $⟨ e^{- β W} ⟩ = e^{- β Δ F}$ reduces to the classical FDT for the work fluctuations.

Proof. Expand the Jarzynski equality to second order in $ϵ$ . The free-energy change is $Δ F = ϵ ⟨ W_{1} ⟩ + ϵ^{2} ⟨ W_{2} ⟩ - (ϵ^{2} /2) β (⟨ W_{1}^{2} ⟩ - ⟨ W_{1} ⟩^{2}) + O (ϵ^{3})$ . The expansion of $e^{- β W}$ to second order gives $$ \langle e^{-\beta W}\rangle = 1 - \beta\langle W\rangle + (\beta^2/2)\langle W^2\rangle + O(\epsilon^3). $$ Matching coefficient by coefficient in $ϵ$ , the order- $ϵ$ terms give $⟨ W_{1} ⟩ = ⟨ W_{1} ⟩$ as an identity, and the order- $ϵ^{2}$ terms give $$ \langle W_2\rangle - \frac{\beta}{2}(\langle W_1^2\rangle - \langle W_1\rangle^2) = \langle W_2\rangle - \beta\langle W_1\rangle\langle W_2\rangle + \beta^2 \langle W_1\rangle^2/2 - \ldots $$ Simplifying, the variance of the work in the slow-driving limit satisfies $$ \langle W^2\rangle - \langle W\rangle^2 = 2 k_B T (\langle W\rangle - \Delta F) + O(\epsilon^3), $$ which is the FDT for the work observable: the variance of $W$ equals the temperature times the dissipated work $⟨ W ⟩ - Δ F$ . The full Jarzynski equality is the all-order resummation of this perturbative FDT identity. $□$

Connections Master

Boltzmann distribution 08.01.03. The FDT presupposes the canonical equilibrium state $ρ_{e q} = e^{- β H} / Z$ , which is the quantum incarnation of the Boltzmann distribution. The detailed-balance relation $S_{B A} (- ω) = e^{- β ℏ ω} S_{A B} (ω)$ at the heart of the FDT proof is a direct consequence of the energy denominator $e^{- β E_{n}} / Z$ in the canonical state. Without the Boltzmann distribution there is no FDT; with it, the FDT identifies the equilibrium fluctuation spectrum with the dissipative response.
Free energy 08.01.04. The free energy $F = - k_{B} T lo g Z$ generates the static susceptibilities as $χ_{A B} (ω = 0) = - \partial^{2} F / \partial f_{A} \partial f_{B} ∣_{f = 0}$ , where $f_{A}, f_{B}$ are conjugate fields coupling to $A$ and $B$ . The FDT then identifies these second derivatives with equilibrium correlations, giving the fluctuation-response formula $χ_{A B} (0) = β (⟨ A B ⟩ - ⟨ A ⟩ ⟨ B ⟩)$ for the static case. Phase transitions are visible as divergences of these susceptibilities, and the FDT translates them into divergences of correlation lengths.
Correlation functions 08.05.02. The spectral function $S_{A B} (ω)$ is the Fourier transform of the dynamical correlation function $⟨ A (t) B (0)⟩$ . The connected correlator $⟨ A B ⟩_{c} = ⟨ A B ⟩ - ⟨ A ⟩ ⟨ B ⟩$ at equal times is the static limit of the integrated spectral function, and the FDT extends the static fluctuation-response identity to its full dynamical form. Critical phenomena live in the divergence structure of correlation functions, and the FDT provides their experimental signatures (relaxation rates, noise spectra, response measurements).
Fokker-Planck equation 08.10.02. The detailed-balance condition that selects the Gibbs distribution as the reversible stationary solution of the Fokker-Planck equation, $J_{e q} = - p_{e q} \nabla V - T \nabla p_{e q} = 0$ , is the classical-limit analogue of the KMS condition that organises the quantum FDT. The Einstein relation $γ D = k_{B} T$ in the Brownian-motion limit of the Fokker-Planck equation is the simplest instance of the FDT, predating the Callen-Welton theorem by 46 years and serving as its prototype.
Onsager solution 08.03.01. Onsager's 1944 transfer-matrix solution of the 2D Ising model gives the equilibrium correlations and susceptibilities exactly at the critical point. The FDT applied to the Ising magnetisation observable predicts the equilibrium magnetisation fluctuations from the experimentally measurable susceptibility, providing a stringent test of the theorem against a fully solved model. Onsager's 1931 reciprocity theorem (a separate but related result) is the static $ω \to 0$ limit of the FDT-implied symmetry of kinetic coefficients.

Historical & philosophical context Master

The historical line begins with Einstein's 1905 paper Über die von der molekularkinetischen Theorie der Wärme geforderte Bewegung von in ruhenden Flüssigkeiten suspendierten Teilchen (Ann. d. Phys. 17, 549) ^{[Einstein 1905]}, which derived the Brownian-motion diffusion coefficient $D = k_{B} T / γ$ from a microscopic balance between thermal jitter and Stokes drag on a suspended particle. Einstein's formula identifies the diffusion coefficient (a fluctuation quantity, characterising spontaneous random walks) with the inverse friction coefficient (a dissipation quantity, characterising the resistance to driven motion), multiplied by the temperature. This is the prototype of the modern FDT: a fluctuation quantity and a dissipation quantity are the same up to a temperature factor. Smoluchowski 1906 (Ann. d. Phys. 21, 756) and Langevin 1908 (C. R. Acad. Sci. 146, 530) refined the framework with the Smoluchowski equation and the Langevin stochastic differential equation respectively, and Perrin's 1909 (Ann. Chim. Phys. 18, 1) experiments confirmed the Einstein relation quantitatively, providing the first direct measurement of Avogadro's number.

The next major step was Nyquist's 1928 paper Thermal agitation of electric charge in conductors (Phys. Rev. 32, 110) ^{[Nyquist 1928]}, which derived the voltage-noise formula $⟨ V^{2} ⟩ = 4 k_{B} T R Δ ν$ for a resistor in thermal equilibrium, using a thermodynamic argument based on the equipartition of energy among the modes of an idealised transmission line terminated by the resistor. Johnson's 1928 companion paper Thermal agitation of electricity in conductors (Phys. Rev. 32, 97) ^{[Johnson 1928]} supplied the experimental confirmation. The Johnson-Nyquist formula is the FDT applied to the voltage observable in a resistor, with $χ_{V V} (ω) = R$ the resistive impedance, and it became the prototype noise formula in electrical engineering and condensed-matter physics.

The modern formulation arrived with the Callen-Welton 1951 paper Irreversibility and generalized noise (Phys. Rev. 83, 34) ^{[Callen-Welton 1951]}, which derived the quantum fluctuation-dissipation theorem $Im χ (ω) = (1/2ℏ) (1 - e^{- β ℏ ω}) S (ω)$ in full generality from the spectral representation of the canonical ensemble. Callen and Welton recognised that the Einstein, Nyquist, and Johnson results were special cases of a single identity governing every observable in a quantum system in thermal equilibrium. Their derivation introduced the modern language of generalised susceptibilities and equilibrium correlation functions, and their formula remains the standard expression of the FDT in textbook treatments.

Kubo's 1957 paper Statistical-mechanical theory of irreversible processes I (J. Phys. Soc. Jpn. 12, 570) ^{[Kubo 1957]} developed linear-response theory in its modern operator-algebraic form. Kubo's expression $χ_{A B} (t) = - (i /ℏ) θ (t) ⟨[A (t), B (0)]⟩$ for the retarded response function as a commutator expectation became the canonical formulation, and the Kubo formula for transport coefficients (electrical conductivity, magnetic susceptibility, viscosity) as Green-Kubo time integrals of equilibrium current-current correlations is the operational tool by which the FDT is applied in condensed-matter physics. Onsager's 1931 Reciprocal relations in irreversible processes (Phys. Rev. 37, 405; 38, 2265) ^{[Onsager 1931]} had already established the symmetry $L_{ij} = L_{j i}$ for static kinetic coefficients from time-reversal invariance, and the Callen-Welton-Kubo framework absorbed Onsager reciprocity as the static limit of the frequency-domain FDT-implied symmetry. Kubo's 1966 review The fluctuation-dissipation theorem (Rep. Prog. Phys. 29, 255) ^{[Kubo 1966]} consolidated the formalism and introduced the second FDT relating noise to friction in a generalised Langevin equation, distinguishing it from the first FDT (Callen-Welton) relating susceptibility to correlator. The two FDTs are equivalent at the level of full information but address different practical questions.

The C*-algebraic reformulation due to Haag-Hugenholtz-Winnink (Comm. Math. Phys. 5, 215, 1967) ^{[Haag-Hugenholtz-Winnink 1967]} promoted the KMS condition to a definition: a state on a $C^{*}$ -dynamical system is in thermal equilibrium at inverse temperature $β$ if and only if it satisfies the KMS condition $ω (A σ_{i β} (B)) = ω (B A)$ on a dense subalgebra of analytic elements. The Bratteli-Robinson monograph Operator Algebras and Quantum Statistical Mechanics 2 ^{[Bratteli-Robinson §5.3]} develops this framework as the rigorous foundation of equilibrium statistical mechanics on infinite quantum systems, with the FDT as a consequence of the KMS condition. Tomita-Takesaki modular theory identifies the analytic continuation $t \to t + i β ℏ$ in the KMS condition with the modular automorphism group $σ_{t}^{ω}$ of the state $ω$ , providing a deeper structural understanding of the FDT in terms of the modular structure of the equilibrium state.

The late-20th-century extension to non-equilibrium statistical mechanics produced the Jarzynski equality (Jarzynski 1997 Phys. Rev. Lett. 78, 2690) ^{[Jarzynski 1997]} and the Crooks fluctuation theorem (Crooks 1999 Phys. Rev. E 60, 2721) ^{[Crooks 1999]}. These are exact identities for the work distribution in driven non-equilibrium processes, containing the FDT as their linear-response limit and recovering the second law as the inequality $⟨ W ⟩ \geq Δ F$ via Jensen's inequality. Stochastic thermodynamics in the sense of Seifert (Seifert 2012 Rep. Prog. Phys. 75, 126001) organises modern small-system thermodynamics around these identities, with the FDT as the linearised governing principle and the Jarzynski-Crooks framework as its non-linear refinement. The single-molecule experiments of Bustamante and collaborators (Liphardt et al. 2002 Science 296, 1832) provided the first quantitative experimental tests of the Jarzynski equality, using optical-trap measurements of RNA hairpin unfolding to verify the identity within the precision of modern biophysical measurement.

Bibliography Master

@book{LandauLifshitzStatPhys1,
  author    = {Landau, L. D. and Lifshitz, E. M.},
  title     = {Statistical Physics, Part 1},
  edition   = {3},
  publisher = {Pergamon Press},
  series    = {Course of Theoretical Physics},
  volume    = {5},
  year      = {1980}
}

@article{CallenWelton1951,
  author  = {Callen, Herbert B. and Welton, Theodore A.},
  title   = {Irreversibility and Generalized Noise},
  journal = {Phys. Rev.},
  volume  = {83},
  pages   = {34--40},
  year    = {1951}
}

@article{Kubo1957,
  author  = {Kubo, Ryogo},
  title   = {Statistical-Mechanical Theory of Irreversible Processes. I},
  journal = {J. Phys. Soc. Jpn.},
  volume  = {12},
  pages   = {570--586},
  year    = {1957}
}

@article{Kubo1966FDT,
  author  = {Kubo, Ryogo},
  title   = {The fluctuation-dissipation theorem},
  journal = {Rep. Prog. Phys.},
  volume  = {29},
  pages   = {255--284},
  year    = {1966}
}

@article{Nyquist1928,
  author  = {Nyquist, H.},
  title   = {Thermal Agitation of Electric Charge in Conductors},
  journal = {Phys. Rev.},
  volume  = {32},
  pages   = {110--113},
  year    = {1928}
}

@article{Johnson1928,
  author  = {Johnson, J. B.},
  title   = {Thermal Agitation of Electricity in Conductors},
  journal = {Phys. Rev.},
  volume  = {32},
  pages   = {97--109},
  year    = {1928}
}

@article{Einstein1905Brownian,
  author  = {Einstein, A.},
  title   = {{\"U}ber die von der molekularkinetischen Theorie der W{\"a}rme geforderte Bewegung von in ruhenden Fl{\"u}ssigkeiten suspendierten Teilchen},
  journal = {Annalen der Physik},
  volume  = {17},
  pages   = {549--560},
  year    = {1905}
}

@article{Onsager1931a,
  author  = {Onsager, Lars},
  title   = {Reciprocal Relations in Irreversible Processes. I.},
  journal = {Phys. Rev.},
  volume  = {37},
  pages   = {405--426},
  year    = {1931}
}

@article{Onsager1931b,
  author  = {Onsager, Lars},
  title   = {Reciprocal Relations in Irreversible Processes. II.},
  journal = {Phys. Rev.},
  volume  = {38},
  pages   = {2265--2279},
  year    = {1931}
}

@article{HaagHugenholtzWinnink1967,
  author  = {Haag, R. and Hugenholtz, N. M. and Winnink, M.},
  title   = {On the equilibrium states in quantum statistical mechanics},
  journal = {Comm. Math. Phys.},
  volume  = {5},
  pages   = {215--236},
  year    = {1967}
}

@book{BratteliRobinson2,
  author    = {Bratteli, Ola and Robinson, Derek W.},
  title     = {Operator Algebras and Quantum Statistical Mechanics 2},
  edition   = {2},
  publisher = {Springer-Verlag},
  year      = {1997}
}

@article{Jarzynski1997,
  author  = {Jarzynski, C.},
  title   = {Nonequilibrium Equality for Free Energy Differences},
  journal = {Phys. Rev. Lett.},
  volume  = {78},
  pages   = {2690--2693},
  year    = {1997}
}

@article{Crooks1999,
  author  = {Crooks, Gavin E.},
  title   = {Entropy production fluctuation theorem and the nonequilibrium work relation for free energy differences},
  journal = {Phys. Rev. E},
  volume  = {60},
  pages   = {2721--2726},
  year    = {1999}
}

@book{KuboTodaHashitsume,
  author    = {Kubo, Ryogo and Toda, Morikazu and Hashitsume, Natsuki},
  title     = {Statistical Physics II: Nonequilibrium Statistical Mechanics},
  publisher = {Springer-Verlag},
  series    = {Springer Series in Solid-State Sciences},
  volume    = {31},
  year      = {1985}
}

@book{Zwanzig2001,
  author    = {Zwanzig, Robert},
  title     = {Nonequilibrium Statistical Mechanics},
  publisher = {Oxford University Press},
  year      = {2001}
}

@book{ChaikinLubensky,
  author    = {Chaikin, Paul M. and Lubensky, Tom C.},
  title     = {Principles of Condensed Matter Physics},
  publisher = {Cambridge University Press},
  year      = {1995}
}

@book{Forster1975,
  author    = {Forster, Dieter},
  title     = {Hydrodynamic Fluctuations, Broken Symmetry, and Correlation Functions},
  publisher = {Benjamin},
  year      = {1975}
}

@article{Seifert2012,
  author  = {Seifert, Udo},
  title   = {Stochastic thermodynamics, fluctuation theorems and molecular machines},
  journal = {Rep. Prog. Phys.},
  volume  = {75},
  pages   = {126001},
  year    = {2012}
}

@article{LiphardtBustamante2002,
  author  = {Liphardt, J. and Dumont, S. and Smith, S. B. and Tinoco Jr., I. and Bustamante, C.},
  title   = {Equilibrium information from nonequilibrium measurements in an experiment of {J}arzynski's equality},
  journal = {Science},
  volume  = {296},
  pages   = {1832--1835},
  year    = {2002}
}

Prerequisites

08.01.01
08.01.03
08.01.04
08.05.02
08.10.02

Tier anchors

beginner: Reif, *Fundamentals of Statistical and Thermal Physics* (McGraw-Hill, 1965), Ch. 15 informal opening on Johnson noise and Brownian motion; Einstein 1905 *Ann. d. Phys.* 17, 549 as the historical narrative
intermediate: Chaikin-Lubensky, *Principles of Condensed Matter Physics* (Cambridge, 1995), Ch. 7 (linear-response theory, dissipation, fluctuation spectra); Forster, *Hydrodynamic Fluctuations, Broken Symmetry, and Correlation Functions* (Benjamin, 1975), Ch. 5; Reichl, *A Modern Course in Statistical Physics*, 4th ed. (Wiley, 2016), Ch. 15
master: Landau-Lifshitz, *Statistical Physics, Part 1*, 3rd ed. (Pergamon, 1980), Vol. 5, §§124-126 (canonical FDT derivation); Callen-Welton, *Phys. Rev.* 83, 34 (1951); Kubo, *J. Phys. Soc. Jpn.* 12, 570 (1957); Kubo-Toda-Hashitsume, *Statistical Physics II: Nonequilibrium Statistical Mechanics* (Springer, 1985), Chs. 2-4; Zwanzig, *Nonequilibrium Statistical Mechanics* (Oxford, 2001), Chs. 1-2; Bratteli-Robinson, *Operator Algebras and Quantum Statistical Mechanics 2*, 2nd ed. (Springer, 1997), §5.3 (KMS condition)

References

Landau, L. D. & Lifshitz, E. M. — Statistical Physics, Part 1, 3rd ed. · Vol. 5 of the Course of Theoretical Physics (Pergamon, 1980), §§124-126: generalised susceptibility, the fluctuation-dissipation theorem, and applications to thermodynamic fluctuations; §123 derives the symmetry properties of the susceptibility tensor
Callen, H. B. & Welton, T. A. — Irreversibility and generalized noise · Phys. Rev. 83, 34 (1951) — the originator paper of the modern fluctuation-dissipation theorem in its full quantum form, deriving the spectrum of equilibrium fluctuations from the imaginary part of the generalised susceptibility
Kubo, R. — Statistical-mechanical theory of irreversible processes I · J. Phys. Soc. Jpn. 12, 570 (1957) — linear-response theory in its modern form: the commutator expression $\chi_{AB}(t) = -(i/\hbar)\theta(t)\langle[A(t), B(0)]\rangle$ for the retarded response function
Nyquist, H. — Thermal agitation of electric charge in conductors · Phys. Rev. 32, 110 (1928) — the originator derivation of the formula $\langle V^2\rangle = 4 k_B T R \Delta\nu$ for voltage noise across a resistor in thermal equilibrium
Johnson, J. B. — Thermal agitation of electricity in conductors · Phys. Rev. 32, 97 (1928) — the originator experimental measurement of resistor voltage noise, companion paper to Nyquist's theoretical derivation
Einstein, A. — Über die von der molekularkinetischen Theorie der Wärme geforderte Bewegung von in ruhenden Flüssigkeiten suspendierten Teilchen · Ann. d. Phys. 17, 549 (1905) — the originator derivation of the Einstein relation $\gamma D = k_B T$ from a microscopic balance between diffusion and Stokes drag on a Brownian particle
Onsager, L. — Reciprocal relations in irreversible processes I, II · Phys. Rev. 37, 405 (1931) and 38, 2265 (1931) — the originator reciprocity theorem for kinetic coefficients $L_{ij} = L_{ji}$ in linear irreversible thermodynamics, proved from microscopic time-reversal symmetry
Kubo, R. — The fluctuation-dissipation theorem · Rep. Prog. Phys. 29, 255 (1966) — review paper containing the second fluctuation-dissipation theorem for stochastic dynamics: the noise kernel of a generalised Langevin equation equals $k_B T$ times the memory friction kernel
Jarzynski, C. — Nonequilibrium equality for free energy differences · Phys. Rev. Lett. 78, 2690 (1997) — the Jarzynski equality $\langle e^{-\beta W}\rangle = e^{-\beta \Delta F}$ as a non-equilibrium identity that recovers FDT in the linear-response limit
Crooks, G. E. — Entropy production fluctuation theorem and the nonequilibrium work relation · Phys. Rev. E 60, 2721 (1999) — the Crooks fluctuation theorem $P_F(W)/P_R(-W) = e^{\beta(W - \Delta F)}$ as a refined non-equilibrium identity that contains the Jarzynski equality and the FDT as special cases
Bratteli, O. & Robinson, D. W. — Operator Algebras and Quantum Statistical Mechanics 2, 2nd ed. · Springer, 1997, §5.3 — the KMS condition $\omega(A \sigma_{i\beta}(B)) = \omega(BA)$ on a C*-dynamical system as the rigorous definition of a thermal equilibrium state on an infinite quantum system, which is the conceptual underpinning of the FDT proof
Chaikin, P. M. & Lubensky, T. C. — Principles of Condensed Matter Physics · Cambridge, 1995, Ch. 7 — linear-response theory, dissipation, Kubo formulas, and the FDT in the condensed-matter setting; Ch. 7 §§7.4-7.5 work out the resistor and viscous-particle examples explicitly
Forster, D. — Hydrodynamic Fluctuations, Broken Symmetry, and Correlation Functions · Benjamin, 1975, Ch. 5 — classic exposition of the FDT in the framework of dynamic correlation functions and the Mori-Zwanziger projection-operator method
Zwanzig, R. — Nonequilibrium Statistical Mechanics · Oxford, 2001, Chs. 1-2 — the projection-operator derivation of generalised Langevin dynamics from a Hamiltonian system, with the second FDT relating noise and friction in the resulting equation of motion
Kubo, R., Toda, M. & Hashitsume, N. — Statistical Physics II: Nonequilibrium Statistical Mechanics · Springer Series in Solid-State Sciences 31, 1985, Chs. 2-4 — the canonical textbook treatment of linear response, Kubo formulas, and the fluctuation-dissipation theorem in its quantum form

Estimated time

beginner: 22m
intermediate: 50m
master: 90m